Local-first memory for AI coding agents
local first

Palace is designed so individual developers can start with local memory before a team commits to a shared self-hosted server.

Works with Cursor, Claude Code, Codex, and other MCP-capable clients.
Combines semantic retrieval, BM25-style search, diaries, and knowledge graph recall.
Open-source MIT core for solo developers; paid server for shared team memory.

Start on your machine

Install palace-rs and give your agent durable memory without creating an account or wiring up a cloud service.

Grow into team memory

When several developers need the same memory layer, Palace Server gives the team a shared deployment while preserving infrastructure control.

Why it matters

Local-first adoption makes it easy to prove the workflow before asking a team to trust a central service.